While pivotal for risk pricing, derivative markets for raw fibers exhibit complex and nonlinear risk transmission mechanisms that challenge conventional monitoring of textile value chains. This study constructs dynamic risk spillover networks using an Elastic-Net-VAR-DY framework and develops a novel GAT-Transformer integrated model to examine risk transmission mechanisms in China's derivative markets. Our findings reveal that risk spillovers exhibit strong timevarying and asymmetric characteristics. Further, the total spillover index (TSI) intensifies sharply during extreme events, such as the COVID-19 pandemic and geopolitical conflicts, resulting in prominent risk resonance. Our empirical evidence identifies systemically important derivatives, such as CSI 300 stock index futures and crude oil futures, and clarifies the key pathways linking them.The proposed model captures spatial and temporal dependencies in risk propagation and significantly outperforms traditional econometric and deep learning benchmarks across nodal risk prediction, contagion intensity measurement, and systemic vulnerability assessment. The study's multi-task learning framework supports a three-tier early warning system targeting nodal risk, contagion intensity, and systemic vulnerability. This approach provides a quantitative foundation for precise and differentiated risk control. These insights emphasize the potential of advanced AI models to enhance financial risk surveillance and can help guide regulators and textile market participants in promoting the stability of fiber-based derivative markets. By integrating these predictive capabilities, stakeholders can better navigate the price volatility inherent in raw material hedging to ensure the long-term resilience of the textile sector.

derivative markets, dynamic risk spillover network, graph attention network, transformer, textile market